Re: unknown partition table starting with 2.6.28

[Bill Davidsen <davidsen@tmr.com>]

Bill Davidsen wrote:
> John Robinson wrote:
>> On 01/02/2010 20:46, Bill Davidsen wrote:
>>> John Robinson wrote:
>>>> On 15/01/2010 23:58, Timothy D. Lenz wrote:
>>>>> I am trying to update my kernel from 2.6.26.8 to the current .32. 
>>>> [...]
>>>>> Starting with .28 I am getting an error about unknown partition 
>>>>> table for all 3 md's. md0 is boot and main programs, md1 is swap, 
>>>>> md2 is mostly recordings storage for vdr. All 3 are raid 1 and 
>>>>> raid is built in.
>>>>
>>>> Your md devices aren't partitioned so you can quite safely ignore 
>>>> the warning. See also 
>>>> http://marc.info/?l=linux-raid&m=125797242110594&w=2
>>>
>>> To clarify that a bit, the kernel can use several partition formats, 
>>> and something in the partitions looks like a partition table but not 
>>> a *valid* partition table. So the kernel warns that it doesn't 
>>> recognize the table.
>>>
>>> I suspect that using a different superblock type would change 
>>> (probably eliminate) this, putting the md information at the start 
>>> of the partition, of in a bit or whatever makes the kernel happy. 
>>> The kernel would make us happy if it checked for a valid md 
>>> superblock at the *end* of the partition, but there may be reasons 
>>> why that's undesirable.
>>>
>>> Finally, I'm less willing than John to say you can ignore it, any 
>>> time something comes close enough to working (in an undesired way) 
>>> to generate an error message, if there's a simple way to be sure the 
>>> kernel doesn't try to use random data as a partition table, you 
>>> might well want to take a step to prevent a problem now.
>>>
>>> I believe it arises out of all arrays being partitionable recently, 
>>> again the details don't come to mid, I've been pretty head down on 
>>> another project since November.
>>
>> I don't think this analysis is correct. Yes, the situation has arisen 
>> out of all arrays - in fact all block devices - being partitionable, 
>> but the warning's not because of something that looks like a dodgy 
>> partition table, it is precisely what it says, a statement that the 
>> device does not contain a valid partition table. I am essentially 
>> repeating the contents of Doug Ledford's earlier post to this list, 
>> to which I referred above.
>
> But the question is, *should* it contain a valid partition table, or 
> even anything which looks enough like a partition table to have the 
> kernel look at it hard enough to think it's invalid? I have several 
> devices on one system which contain essentially random data, and I 
> don't see this, so I assume that my data never looks enough like a 
> partition table to trigger this. At least to 2.6.33-rc6, which I did 
> boot.
>
On reflection I may not have said this clearly. I have block devices 
which do not have partition tables and which do not trigger this 
message. Therefore something is triggering this message, beyond the lack 
of a partition table. My thought is that it may be some logic called 
when the array is assembled, and some data on the array.

-- 
Bill Davidsen <davidsen@tmr.com>
  "We can't solve today's problems by using the same thinking we
   used in creating them." - Einstein